Refactor SLO workload to use environment variables

.github/workflows/slo-report.yml

@@ -16,42 +16,7 @@ jobs:
       pull-requests: write
     steps:
       - name: Publish YDB SLO Report
-        uses: ydb-platform/ydb-slo-action/report@13c687b7d4b2879da79dd12932dee0ed2b65dd1c
+        uses: ydb-platform/ydb-slo-action/report@v2
         with:
           github_token: ${{ secrets.GITHUB_TOKEN }}
           github_run_id: ${{ github.event.workflow_run.id }}
-
-  remove-slo-label:
-    if: always() && github.event.workflow_run.event == 'pull_request'
-    name: Remove SLO Label
-    needs: ydb-slo-action-report
-    runs-on: ubuntu-latest
-    permissions:
-      pull-requests: write
-    steps:
-      - name: Remove SLO label from PR
-        uses: actions/github-script@v7
-        with:
-          script: |
-            const pullRequests = context.payload.workflow_run.pull_requests;
-            if (pullRequests && pullRequests.length > 0) {
-              for (const pr of pullRequests) {
-                try {
-                  await github.rest.issues.removeLabel({
-                    owner: context.repo.owner,
-                    repo: context.repo.repo,
-                    issue_number: pr.number,
-                    name: 'SLO'
-                  });
-                  console.log(`Removed SLO label from PR #${pr.number}`);
-                } catch (error) {
-                  if (error.status === 404) {
-                    console.log(`SLO label not found on PR #${pr.number}, skipping`);
-                  } else {
-                    throw error;
-                  }
-                }
-              }
-            } else {
-              console.log('No pull requests associated with this workflow run');
-            }

.github/workflows/slo.yml

@@ -42,16 +42,14 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        include:
-          - id: sync-table
-            prefix: table
-            workload: sync-table
-          - id: sync-query
-            prefix: table
-            workload: sync-query
+        sdk:
+          - name: sync-table
+            command: "--read-rps ${{ inputs.slo_workload_read_max_rps || '1000' }} --write-rps ${{ inputs.slo_workload_write_max_rps || '100' }}"
+          - name: sync-query
+            command: "--read-rps ${{ inputs.slo_workload_read_max_rps || '1000' }} --write-rps ${{ inputs.slo_workload_write_max_rps || '100' }}"
 
     concurrency:
-      group: slo-${{ github.ref }}-${{ matrix.workload }}
+      group: slo-${{ github.ref }}-${{ matrix.sdk.name }}
       cancel-in-progress: true
 
     steps:
@@ -84,7 +82,7 @@ jobs:
           docker compose version
 
       - name: Checkout current version
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
         with:
           path: current
           fetch-depth: 0
@@ -118,16 +116,10 @@ jobs:
           echo "ref=$BASELINE_REF" >> $GITHUB_OUTPUT
 
       - name: Checkout baseline version
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
         with:
           ref: ${{ steps.baseline.outputs.sha }}
           path: baseline
-          fetch-depth: 1
-
-      - name: Show Docker versions
-        run: |
-          docker --version
-          docker compose version
 
       - name: Build workload images (current + baseline)
         run: |
@@ -141,125 +133,32 @@ jobs:
             -t "ydb-app-baseline" \
             "$GITHUB_WORKSPACE/baseline"
 
-      - name: Initialize YDB SLO
-        id: ydb_slo
-        uses: ydb-platform/ydb-slo-action/init@13c687b7d4b2879da79dd12932dee0ed2b65dd1c
+      - name: Run SLO Tests
+        uses: ydb-platform/ydb-slo-action/init@v2
+        timeout-minutes: 30
         with:
-          github_issue: ${{ github.event.pull_request.number || inputs.github_issue }}
+          github_issue: ${{ github.event.inputs.github_issue }}
           github_token: ${{ secrets.GITHUB_TOKEN }}
-          workload_name: ydb-python-${{ matrix.workload }}
+          workload_name: ${{ matrix.sdk.name }}
+          workload_duration: ${{ inputs.slo_workload_duration_seconds || '600' }}
           workload_current_ref: ${{ github.head_ref || github.ref_name }}
+          workload_current_image: ydb-app-current
+          workload_current_command: ${{ matrix.sdk.command }}
           workload_baseline_ref: ${{ steps.baseline.outputs.ref }}
-
-      - name: Prepare SLO Database
-        run: |
-          docker run --rm \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "WORKLOAD=${{ matrix.workload }}" \
-            -e "REF=${{ github.head_ref || github.ref_name }}" \
-            ydb-app-current \
-            ${{ matrix.prefix }}-create grpc://ydb:2136 /Root/testdb
-
-      - name: Run SLO Tests (current + baseline in parallel)
-        timeout-minutes: 15
-        env:
-          WORKLOAD: ${{ matrix.workload }}
-          DURATION: ${{ inputs.slo_workload_duration_seconds || 600 }}
-          READ_RPS: ${{ inputs.slo_workload_read_max_rps || 1000 }}
-          WRITE_RPS: ${{ inputs.slo_workload_write_max_rps || 100 }}
-          CURRENT_REF: ${{ github.head_ref || github.ref_name }}
-          BASELINE_REF: ${{ steps.baseline.outputs.ref }}
-        run: |
-          ARGS="${{ matrix.prefix }}-run grpc://ydb:2136 /Root/testdb \
-            --otlp-endpoint http://prometheus:9090/api/v1/otlp/v1/metrics \
-            --report-period 250 \
-            --time ${DURATION} \
-            --read-rps ${READ_RPS} \
-            --write-rps ${WRITE_RPS} \
-            --read-timeout 1000 \
-            --write-timeout 1000"
-
-          echo "Starting current workload (ref=${CURRENT_REF}, workload=${WORKLOAD})..."
-          docker run -d \
-            --name ydb-app-current \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "REF=${CURRENT_REF}" \
-            -e "WORKLOAD=${WORKLOAD}" \
-            ydb-app-current \
-            $ARGS
-
-          echo "Starting baseline workload (ref=${BASELINE_REF}, workload=${WORKLOAD})..."
-          docker run -d \
-            --name ydb-app-baseline \
-            --network ydb_ydb-net \
-            --add-host "ydb:172.28.0.11" \
-            --add-host "ydb:172.28.0.12" \
-            --add-host "ydb:172.28.0.13" \
-            --add-host "ydb:172.28.0.99" \
-            -e "REF=${BASELINE_REF}" \
-            -e "WORKLOAD=${WORKLOAD}" \
-            ydb-app-baseline \
-            $ARGS
-
-          echo ""
-          echo "==================== INITIAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== INITIAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          echo "Waiting for workloads to complete (${DURATION}s)..."
-          sleep ${DURATION}
-
-          echo "Stopping containers after ${DURATION}s..."
-          docker stop --timeout=30 ydb-app-current ydb-app-baseline 2>&1 || true
-
-          # Force kill if still running
-          docker kill ydb-app-current ydb-app-baseline 2>&1 || true
-
-          # Check exit codes
-          CURRENT_EXIT=$(docker inspect ydb-app-current --format='{{.State.ExitCode}}' 2>/dev/null || echo "1")
-          BASELINE_EXIT=$(docker inspect ydb-app-baseline --format='{{.State.ExitCode}}' 2>/dev/null || echo "1")
-
-          echo "Current exit code: ${CURRENT_EXIT}"
-          echo "Baseline exit code: ${BASELINE_EXIT}"
-
-          echo ""
-          echo "==================== FINAL CURRENT LOGS ===================="
-          docker logs -n 15 ydb-app-current 2>&1 || echo "No current container"
-          echo ""
-          echo "==================== FINAL BASELINE LOGS ===================="
-          docker logs -n 15 ydb-app-baseline 2>&1 || echo "No baseline container"
-          echo ""
-
-          if [[ "${CURRENT_EXIT}" != "0" || "${BASELINE_EXIT}" != "0" ]]; then
-            echo "One or both workloads failed."
-            exit 0
-          fi
-
-          echo "SUCCESS: Workloads completed successfully"
-
-      - if: always()
-        name: Store logs
-        run: |
-          docker logs ydb-app-current > current.log 2>&1 || echo "No current container" > current.log
-          docker logs ydb-app-baseline > baseline.log 2>&1 || echo "No baseline container" > baseline.log
-
-      - if: always()
-        name: Upload logs
-        uses: actions/upload-artifact@v4
+          workload_baseline_image: ydb-app-current
+          workload_baseline_command: ${{ matrix.sdk.command }}
+
+  ydb-slo-action-report:
+    runs-on: ubuntu-latest
+    name: Publish YDB SLO Report
+    needs: ydb-slo-action
+    permissions:
+      checks: write
+      contents: read
+      pull-requests: write
+    steps:
+      - name: Publish YDB SLO Report
+        uses: ydb-platform/ydb-slo-action/report@v2
         with:
-          name: ydb-python-${{ matrix.workload }}-logs
-          path: |
-            ./current.log
-            ./baseline.log
-          retention-days: 1
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          github_run_id: ${{ github.run_id }}

AGENTS.md

@@ -90,44 +90,47 @@ To regenerate protobuf stubs: see `Makefile` and `generate-protobuf.Dockerfile`.
 
 ---
 
-## Topic Chaos Testing (SLO)
+## SLO Testing
 
-Run this only for changes that affect topic reader/writer reconnection logic.
+Run this for changes that affect topic/table reader/writer reconnection logic.
 
-**1. Start YDB with chaos** (kills a DB node every ~20 seconds):
-```sh
-docker compose -f tests/slo/playground/configs/compose.yaml up -d
-```
+### Docker Compose (full stack)
 
-**2. Wait until YDB is healthy:**
-```sh
-docker ps --format "table {{.Names}}\t{{.Status}}" | grep ydb
-```
+Uses [ydb-slo-action](https://github.com/ydb-platform/ydb-slo-action/tree/v2/deploy) infra (YDB cluster + Prometheus + workload in one command).
 
-**3. Create a test topic** (from `tests/slo/` directory):
+From `tests/slo/` directory:
 ```sh
-source .venv/bin/activate
-python ./src topic-create grpc://localhost:2135 /Root/testdb \
-    --path /Root/testdb/slo_topic --debug
+WORKLOAD_NAME=topic ./slo_runner.sh
+WORKLOAD_NAME=sync-query ./slo_runner.sh
 ```
 
-**4. Test writer** (60 sec):
+Override defaults via env vars: `RUN_TIME_SEC`, `WRITE_RPS`, `READ_THREADS`, `WRITE_THREADS`, `MESSAGE_SIZE`, `DEBUG=1`.
+
+### Local run (against your own YDB)
+
+**1. Start playground cluster:**
 ```sh
-python ./src topic-run grpc://localhost:2135 /Root/testdb \
-    --path /Root/testdb/slo_topic --otlp-endpoint "" \
-    --read-threads 0 --write-rps 1 --time 60 --debug
+docker compose -f tests/slo/playground/configs/compose.yaml up -d
 ```
 
-**5. Test reader** (60 sec):
+**2. Run workload** (from `tests/slo/` directory):
 ```sh
-python ./src topic-run grpc://localhost:2135 /Root/testdb \
-    --path /Root/testdb/slo_topic --otlp-endpoint "" \
-    --read-rps 1 --write-threads 0 --time 60 --debug
+source ../../.venv/bin/activate
+
+# Topic workload (60 sec)
+python ./src grpc://localhost:2136 /Root/testdb \
+    --workload-name topic --otlp-endpoint "" --time 60 --debug
+
+# Table workload (60 sec)
+python ./src grpc://localhost:2136 /Root/testdb \
+    --workload-name sync-query --otlp-endpoint "" --time 60 --debug
 ```
 
-**6. Tear down:**
+**3. Tear down:**
 ```sh
 docker compose -f tests/slo/playground/configs/compose.yaml down
 ```
 
-**Success criteria:** writer and reader reconnect automatically during node restarts with no fatal errors.
+Full list of CLI arguments and environment variables: see `tests/slo/README.md` or run `python tests/slo/src --help`.
+
+**Success criteria:** workload reconnects automatically during node restarts with no fatal errors.

pyproject.toml

@@ -1,3 +1,9 @@
+[tool.ty.environment]
+extra-paths = ["tests/slo/src"]
+
+[tool.ruff]
+line-length = 120
+
 [tool.black]
 line-length = 120
 

tests/slo/.gitignore

@@ -0,0 +1 @@
+.infra/

tests/slo/Dockerfile

@@ -1,38 +1,42 @@
 # syntax=docker/dockerfile:1
 
 # This image packages the Python SLO workload runner.
-# It expects to be run with arguments like:
-#   docker run --rm <image> table-run <endpoint> <db> --otlp-endpoint http://prometheus:9090/api/v1/otlp/v1/metrics ...
+#
+# All settings are CLI arguments with env-var fallback (CLI arg > env var > default):
+#   endpoint              $YDB_ENDPOINT                 grpc://ydb:2136
+#   db                    $YDB_DATABASE                 /Root/testdb
+#   --workload-name       $WORKLOAD_NAME                sync-query
+#   --workload-ref        $WORKLOAD_REF / $REF          main
+#   --otlp-endpoint       $OTEL_EXPORTER_OTLP_ENDPOINT  http://ydb-prometheus:9090/api/v1/otlp
+#   --time                $WORKLOAD_DURATION            600
+#
+# Example:
+#   docker run --rm <image> grpc://ydb:2136 /Root/testdb --workload-name topic --time 120
 #
 # Notes:
 # - OpenTelemetry 1.39.x requires Python >= 3.9.
-# - The entrypoint is `python ./tests/slo/src`, i.e. it runs the `__main__.py`
-#   from that directory (same as `python tests/slo/src ...` in CI).
 
-FROM python:3.11-slim AS build
+FROM python:3.11-slim
 
 ENV PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1
 
-WORKDIR /src
-COPY . /src
-
-# Install runtime deps into an isolated venv so we can copy it into the final stage.
-RUN python -m venv /opt/venv \
-    && /opt/venv/bin/python -m pip install --no-cache-dir --upgrade pip \
-    && /opt/venv/bin/pip install --no-cache-dir . \
-    && /opt/venv/bin/pip install --no-cache-dir -r tests/slo/requirements.txt
+RUN apt-get update && apt-get install -y --no-install-recommends gcc libc6-dev && rm -rf /var/lib/apt/lists/*
 
+WORKDIR /src
 
-FROM python:3.11-slim
+# 1. YDB SDK
+COPY setup.py pyproject.toml README.md requirements.txt ./
+COPY ydb/ ydb/
+RUN pip install --no-cache-dir .
 
-ENV PYTHONDONTWRITEBYTECODE=1 \
-    PYTHONUNBUFFERED=1 \
-    PATH="/opt/venv/bin:${PATH}"
+# 2. SLO deps
+COPY tests/slo/requirements.txt tests/slo/requirements.txt
+RUN pip install --no-cache-dir -r tests/slo/requirements.txt
 
-WORKDIR /app
-
-COPY --from=build /opt/venv /opt/venv
-COPY --from=build /src/tests/slo/src /app/tests/slo/src
+# 3. Workload source
+COPY tests/slo/src /src/tests/slo/src
 
 ENTRYPOINT ["python", "./tests/slo/src"]
+
+CMD ["--read-rps", "1000", "--write-rps", "100"]